understand
why things
happen.
Ask any question. Get a structured cause-and-effect graph built by Grok AI — events, edges, confidence scores, all connected.
Four steps
from question
to graph.
A three-phase retrieval pipeline ensures AI generation only runs when truly needed — keeping latency low and quality high.
Ask a Question
Type any natural-language question — geopolitical, economic, scientific, or social. No special syntax needed.
Three-Phase Retrieval
Neo4j keyword search runs first. On miss, Qdrant semantic search (≥ 0.70 similarity). Only a true miss triggers Grok AI generation.
Grok Builds the Graph
A structured prompt to xAI's Grok returns a validated casual JSON graph — event nodes with dates, domains, confidence scores, source URLs, and directed edges with perspectives.
Persist & Cache
New nodes are bulk-written to Neo4j, embedded by Ollama (nomic-embed-text), and stored in Qdrant. Redis caches by topic + mode + count. Every query makes the system smarter.
Everything needed to map casuality.
Grok AI Graph Generation
A single structured prompt to xAI's Grok produces a complete casual knowledge graph — nodes with dates, domains, confidence scores, source URLs, and directed edges. No Wikipedia scraping.
Three Generation Modes
Minimal (grok-3-mini, 3k tokens) · Balanced (grok-3, 4k) · Quality (grok-3, 6k). Controllable per request.
Semantic + Keyword Search
Neo4j keyword → Qdrant vector similarity (≥ 0.70) → Grok generation. Only true misses trigger AI.
Casual Perspectives
Edges carry perspective labels and an is_contested flag. See who says what, and how disputed it is.
What-If Simulation
Remove an event from the graph and watch its causal impact recalculate in real time — undo anytime.
One endpoint. Full casual graph.
GET /api/v1/EventNodes/search?q=...&mode=balanced&count=10Built to understand
the world casually.
Graph Sight is a production-grade API system that turns natural-language questions into structured casual knowledge graphs. It combines a .NET 8 backend, a Python FastAPI AI sidecar, and five backing services — Neo4j, Qdrant, Ollama, Redis, and PostgreSQL — all orchestrated in Docker.
Most information systems tell you what happened. Graph Sight tells you why. Every event node has a date, domain, confidence score, and source URL. Every edge has a perspective, strength, and contested flag.
Why, not just what
Most search returns facts. Graph Sight returns causal chains — structured graphs that answer why an event happened, not just that it happened.
Generation as last resort
AI generation is expensive and slow. Three search phases run first — only a genuine cache miss triggers Grok. The system gets smarter with every query.
Confidence over certainty
Every node carries a confidence score (0.4–0.95) and every edge can be marked contested. Graph Sight acknowledges what is disputed rather than flattening it.